Skip to content
View Amadeus415's full-sized avatar
2 followers · 4 following
  • Autodesk
  • San Francisco
  • 09:18 (UTC -07:00)

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Block or report Amadeus415

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Amadeus415/README.md

Hi, I'm Amadeus.

I'm a data engineer at Autodesk in San Francisco. Here is a few side projects.

Project What it is Measured
vendormatch Entity resolution for messy vendor records. Cheap matchers take the easy pairs; a local LLM only sees the hard ones. F1 0.936 vs 0.906 for the best string match. A local Qwen3-14B on 3.6% of pairs cuts false matches 37%. 50k records in 31 s.
Nightshift A prediction-market paper fund that agents run and improve every day. Python holds the money. 96 autonomous cycles, 885 risk checks, 0 breaches
NanoServe A readable LLM serving stack for Apple Silicon: OpenAI-compatible API, streaming, benchmarks. Continuous batching: 2.2x throughput, 4.5x lower worst-case TTFT for 8 Qwen3-14B streams. About 5 ms of server overhead.
NanoTrain Train a small language model from scratch, then serve it with NanoServe. 4.9M-parameter model to 1.56 nats/byte on Tiny Shakespeare in 11 minutes

amadeusc.com · LinkedIn

Pinned Loading

  1. AUTOQUANT AUTOQUANT Public

    Autoresearch for crypto alpha, run as a model arena

    Python

  2. agentic_trading agentic_trading Public

    Nightshift: a prediction-market paper fund that agents run and improve every day. An LLM researches Polymarket and Kalshi; deterministic Python holds the money.

    Python

  3. nanoclaw nanoclaw Public

    Forked from nanocoai/nanoclaw

    A lightweight alternative to OpenClaw that runs in containers for Security. Forked from NanoClaw and rebuilt to run directly on QWEN Code CLI.

    TypeScript

  4. nanoserve nanoserve Public

    Minimal hackable LLM serving stack for learning inference engineering (Qwen3.8-27B on Apple Silicon)

    Python